Self-Confessed Fabrications and Rise of TBDs

Some submissions to ICLR 2027 and papers on arXiv contain a very interesting instruction: to replace the reported experimental results with actual measurements before submission. The manuscripts themselves claim that some results reported in the paper are hypothetical. Here are some examples which I first found in an automated fashion, after which I manually verified and excluded many ambiguous matches. In this process, I also saw another interesting phenomenon: use of “TBD”s and blank placeholders in submitted papers. I detail these observations from ICLR 2027 (papers 1 to 20) and then from arXiv (papers 21 to 40). There seem to be more such papers on arXiv but I did not have the time to manually verify them. Note that these observations are about statements in the papers and do not establish the underlying experiments were conducted, or if AI was involved, or if any intentional misrepresentation occurred.

1. “All values are simulated placeholder results and should be replaced by actual experimental outputs.” Statements like this in Tables 1, 2, 3, 4, 5.

2. “These values are predicted placeholders and must be replaced by measured results before submission”

“Table S3 contains provisional predicted values to define the reporting format; all entries must be replaced with measured human-study results before submission.”

3. ” All numerical results are hypothetical and must be replaced with actual experimental results.”

Table 1 of https://openreview.net/pdf?id=mViu5h9uNN

4. “Novelty values other than PMI are synthetic placeholders; their rankings are illustrative.”

Tables 2 and 3 of https://openreview.net/pdf?id=MWreo6RbeX

5. Table 7: “The non-default quality values and timing entries are reference values for the planned capacity study and should be replaced with measured results.”

“Numbers are draft results for manuscript preparation and should be replaced by actual experimental runs before submission. Best results are in bold.” and similar statements

7. “Values marked with † are proportional projections used only to complete the planned comparison and are not measured results. They must be replaced by matched-protocol measurements before submission.”

8. “Replace these values with measured results.”

Table 7, 8, 11 and 12 of https://openreview.net/pdf?id=dhmtivoA8R

9. “Placeholder for Figure 1: benchmark overview and evaluation pipeline.”

10. “Other rows retain the original draft results; SparseForge entries remain pending sparsity/export verification.” in Table 2. The intended meaning of this statement is unclear.

11. Table 2: “replace the displayed pilot values with the final repeated-run estimates.” Regarding Table 1: “The numerical values in the REPO rows are provisional placeholders for the first complete training run.” Table 3: “entries are pilot values to be replaced by final repeated runs.” Figure 6: “Rare categories are aggregated, and the percentages must be replaced by the final census.”

12. Table 7: ” Numbers are illustrative placeholders for manuscript drafting.”

Table 8: “Numbers are illustrative placeholders.”

13. Numerous placeholders throughout the paper, including in the abstract.

14. “Blue zero standard deviations denote cited results with unreported uncertainty; they are placeholders, not measured deviations.” Table 1 and 13

15. Table 2 and lots of other places in the paper are TBD.

16. “The NavFoM result is still pending; TBD denotes a pending result.”

17. Table 5 and 6: “TBD denotes pending category scores.”

Despite this statement, I could not find any TBDs in the manuscript.

19. Abstract: “[TBD: principal findings, effect sizes, confidence intervals, and scope.]”

20. “Table 5: Model and sampling settings used in the experiments. Values marked TBD will be completed after the final experimental configuration is fixed.”

Most of the table is TBD. Similar issues in Tables 6, 7, 8, 9.

We have shared these observations with the ICLR 2027 organizers.

21. The abstract claims: “Under the same backbone model, trajectory sampling budget, and test-time compute, our method achieves proof success rates of 100.0%, 90.6%, 4/6, and 4/6, respectively, with GPT-5.5, outperforming the baseline methods”

These numbers come from Table 2. But Table 2 as well as Tables 3, 4, 5, 6 have captions of the form “Planning scenario, not measured results”.

22. Section 4.4 on Main Results, Tables 1, 2, 3: “numbers are draft placeholders for writing; replace with your real runs later” or similar statements.

23. Regarding Table 4 “note: the values shown are placeholders and should be replaced with your actual experimental results”

24. Table 6: “All values are simulated placeholders, not experimental results, and must be replaced before submission”

25. Table 3: “All numbers are placeholders and will be replaced with actual results.”

26. Table 1: “Numbers in red are placeholders to be populated from the experimental sweep” All the numbers in the table are red. Table 2: “Numbers are placeholders pending experimental runs.”

27. “(Camera-ready: numerical entries below are placeholders pending final ImageNet runs; the experimental protocol and configurations are fixed.)”

28. Table 3: “Numbers are illustrative placeholders; the final paper will report bootstrap CIs.” Figures 2, 3, 4, 5, 6: “Illustrative results pending experimental replication.”

29. Table 4: “The numbers are illustrative placeholders and must be replaced with measured values.”

30. Table 13: “Numbers are placeholders and will be replaced with final evaluation results.”

31. Table IV: “Numbers shown are placeholders.”

32. Figure 8, Table 2, 3, 5: “All plotted values are placeholders pending measurement” or similar statements.

33. About Table 3: “The behavior-guided TCN and PatchTST results are currently illustrative placeholders awaiting experimental verification.”

34. Tables 1, 2, 3, 4, 5, 6 “[PLACEHOLDER]”

35. Table 5: ““sim” denotes synthetic layout placeholders.” Two out of the three columns have all rows marked with “sim.”

36. “Table II and all plots in this section contain simulated values constructed to test reporting, plotting, and consistency. They are not outcomes of an executed backtest.”

37. Tables 5, 6, 7: “Illustrative values only. No experiments were run.” Table 7: “All performance cells must be replaced.”

38. “All entries except the dense baseline are literature-calibrated estimates rather than directly measured results, and should be replaced by empirical values in the final paper.”

39. Figure 10: “Illustrative; replace with measured performance grid for final submission”

40. Table 2: “Traffic shares and cell assignments are illustrative placeholders pending full benchmark completion“

What is going on? Here are some plausible explanations for this observed phenomenon:

(A) It may be a combination of the following two known phenomenon:

– AI agents conducting autonomous research can fabricate results https://www.pnas.org/doi/10.1073/pnas.2610214123; sometimes they do so simply as placeholders.

– Human authors may not be familiar with the papers they are submitting https://blog.tmlr.org/2026/asking-authors-about-their-own-papers/.

(B) Authors could have forgotten to remove the placeholder statements before submission (irrespective of whether the experiments were actually conducted or not).

(C) These numbers are truly meant to be illustrative of what such a table or figure would look like.

(D) These are synthetic simulations, but where the simulations are reported in a non-traditional way.

These possibilities are not mutually exclusive, and the examples presented here do not provide sufficient evidence to determine which explanation applies to any particular manuscript.